Ruchay Alexey, Gritsenko Svetlana, Ermolova Evgenia, Bochkarev Alexander, Ermolov Sergey, Guo Hao, Pezzuolo Andrea
Federal Research Centre of Biological Systems and Agro-Technologies of the Russian Academy of Sciences, 460000 Orenburg, Russia.
Department of Mathematics, Chelyabinsk State University, 454001 Chelyabinsk, Russia.
Animals (Basel). 2022 Apr 29;12(9):1152. doi: 10.3390/ani12091152.
Live weight is an important indicator of livestock productivity and serves as an informative measure for the health, feeding, breeding, and selection of livestock. In this paper, the live weight of pig was estimated using six morphometric measurements, weight at birth, weight at weaning, and age at weaning. This study utilised a dataset including 340 pigs of the Duroc, Landrace, and Yorkshire breeds. In the present paper, we propose a comparative analysis of various machine learning methods using outlier detection, normalisation, hyperparameter optimisation, and stack generalisation to increase the accuracy of the predictions of the live weight of pigs. The performance of live weight prediction was assessed based on the evaluation criteria: the coefficient of determination, the root-mean-squared error, the mean absolute error, and the mean absolute percentage error. The performance measures in our experiments were also validated through 10-fold cross-validation to provide a robust model for predicting the pig live weight. The StackingRegressor model was found to provide the best results with an MAE of 4.331 and a MAPE of 4.296 on the test dataset.
活重是家畜生产力的重要指标，也是衡量家畜健康、饲养、繁殖和选育的重要依据。本文利用出生体重、断奶体重和断奶年龄这六项形态测量数据对猪的活重进行了估计。本研究使用了一个包含340头杜洛克、长白和约克夏品种猪的数据集。在本文中，我们提出了一种比较分析各种机器学习方法的方法，该方法使用异常值检测、归一化、超参数优化和堆叠泛化来提高猪活重预测的准确性。基于判定系数、均方根误差、平均绝对误差和平均绝对百分比误差等评估标准对活重预测性能进行了评估。我们实验中的性能指标也通过10折交叉验证进行了验证，以提供一个稳健的猪活重预测模型。在测试数据集上，StackingRegressor模型被发现提供了最佳结果，MAE为4.331，MAPE为4.296。
